Trying to understand OTP apps a little better and distribution

Chat / Discussions
, ,
1

I was thinking:

What if I could abstract some of my otp applications from in my phoenix app and move them to another server?

I got this idea because I wanted to see if I could not put one of my 3 free 256mb VMs from fly.io to use since I was hitting a out of memory error.

[info][   41.509215] Out of memory: Killed process 255 (beam.smp) total-vm:1880020kB, anon-rss:141076kB, file-rss:0kB, shmem-rss:69636kB, UID:65534 pgtables:624kB oom_score_adj:0
[error]could not make HTTP request to instance: connection closed before message completed
[info][os_mon] cpu supervisor port (cpu_sup): Erlang has closed
[info][os_mon] memory supervisor port (memsup): Erlang has closed
[info] INFO Main child exited with signal (with signal 'SIGKILL', core dumped? false)
[info] INFO Starting clean up.
[info] WARN hallpass exited, pid: 256, status: signal: 15 (SIGTERM)
[info] INFO Process appears to have been OOM killed!
[info]2023/08/16 00:12:28 listening on .....
[info][   43.303638] reboot: Restarting system
[info]Out of memory: Killed process

So say I have an app I want to break apart. I’m gonna go with libcluster for this example.

application.ex on server A

 def start(_type, _args) do
    topologies = [
      cluster_example: [
        strategy: Cluster.Strategy.Epmd,
        config: [
          hosts: [
            :"app@foo.dev",
            :"app@bar.dev"
          ]
        ]
      ]
    ]

    children = [
      # Start the Ecto repository
      ClusterExample.Repo,
      # Start the Telemetry supervisor
      ClusterExampleWeb.Telemetry,
      # Start a worker by calling: ClusterExample.Worker.start_link(arg)
      # {ClusterExample.Worker, arg}
      {Cluster.Supervisor, [topologies, [name: ClusterExample.ClusterSupervisor]]}
    ]

And then you got server B

 def start(_type, _args) do
    topologies = [
      cluster_example: [
        strategy: Cluster.Strategy.Epmd,
        config: [
          hosts: [
            :"app@foo.dev",
            :"app@bar.dev"
          ]
        ]
      ]
    ]

    children = [
      # Start the PubSub system
      {Phoenix.PubSub, name: ClusterExample.PubSub},
      # Start the Endpoint (http/https)
      ClusterExampleWeb.Endpoint,
      # Start a worker by calling: ClusterExample.Worker.start_link(arg)
      # {ClusterExample.Worker, arg}
      {Cluster.Supervisor, [topologies, [name: ClusterExample.ClusterSupervisor]]}
    ]

Would this work like I assume it would? In that I assume the app will route methods correctly to the given hosted apps?

I look at :observer.start() and I assume I should be able to move any of these apps to another server correct?

image
image612×1746 41.8 KB

I guess I’m just not sure what that would look like.

2 Likes
2

I’m very interested in what others have to say about this.

I’ve been pondering similar questions in the context of a phoenix app that also uses Membrane framework: How to run certain applications or supervisors in certain machines. In this scenario, machine requirements are very different.

:eyes:

1 Like
3

My first thought is this will fail as a result of clashing namespaces. IE: ClusterExample.

I assume at first I would have to update the name spaces in my apps where calling into the given modules.

But I have not tried this.

4

That list of applications is for production node or local node?

5

I’m not really sure what you mean by the question.

Do you mean if I’m running it via a bin vs mix?

Also this was created in the context of this. GitHub - joshchernoff/cluster_example: A very basic example of using distributed elixir with LiveView and docker-compose
That said my version of alpine is crying because of the older elixir 1.11 in that repo so take it with a grain of salt.